A crowded station, a useless 5G signal, and a love for rabbit holes. LLMs offer a unparalleled opportunity to learn new things immediately and in ways that match your preferred learning style.
Track AI agent traffic that bypasses your web analytics. Stream Vercel Log Drains into MotherDuck to see how ChatGPT, Claude, and other agents browse your site.
Build a real-time voting app on Cloudflare Workers that queries MotherDuck through the Postgres endpoint, using Durable Objects for live state and the pg package for analytical workloads.
Measuring the rate of software output growth in the AI era, across Show HN, GitHub, package registries, and research.
Learn when and how to use dbt's microbatch incremental strategy with DuckDB. Covers row groups vs partitions, benchmarks comparing full refresh, merge, delete+insert, and microbatch strategies, plus configuration tips and common pitfalls.
I processed 60,000 job postings with an LLM for $10. That's not a typo. This post walks through how to use OpenAI's Batch API for bulk document processing: building JSONL request files, orchestrating without losing your mind, controlling costs with token limits, and tracking batch state in a database. When AI becomes this cheap, it changes what's possible.
I'll admit it: as a data engineer I still use Excel. But DuckDB has become my secret weapon for ad-hoc data exploration. No more struggling with CSV imports, no more manual column matching across files, no more giving up on JSON. This guide shows you how to query remote CSVs, handle schema drift across multiple files, unnest nested data, and flatten JSON APIs—all with SQL you can remember if you can remember Excel formulas.
If you've inherited a bucket full of thousands of tiny JSON files—one per API call, one per event, one per log minute—you know the pain: slow scans, schema anxiety, and rising warehouse bills. This guide shows you how to consolidate them into clean Parquet with DuckDB: handling schema drift, maintaining lineage, optimizing performance, and integrating with dbt. Touch your raw files once, then model against something stable.
I got tired of copy-pasting articles into ChatGPT for summaries—and even more tired of the bland results. So I built a CLI tool that combines web scraping with LLM personas. Now I can ask Nietzsche what he thinks of Hacker News, or get my skeptical analytics engineer persona to tear apart a vendor blog post. This post shows you how to build faitch: web scraping meets AI characters.
What if dbt could deploy a dashboard instead of just another table? I was tired of switching between my dbt transformations and Streamlit visualization code, so I built a custom materialization that deploys Streamlit apps directly on Snowflake. This tutorial walks through the architecture: using dbt configs, uploading files to stages, and creating apps that know about your data models, lineage, and freshness.